average.fit.models {mmodely}

Calculate a weighted average of pglm

Description

These function takes the output of pgls.iter and uses its list of objects model fits, optimzations (e.g.AICc) and performs a weighted average on the ctoefficients estimated in the former by weighting by the latter. The parameters can also optionally be converted to binary by specifying "binary=FALSE" or just running the alias wraper function for assessing evidence of variable importance (Burnham & Anderson 2000).

Usage

average.fit.models(vars, fits, optims,weight='AICw',
         by=c('n','q','nXq','rwGsm')[1], round.digits=5, binary=FALSE, standardize=FALSE)
variable.importance(vars, fits, optims,weight='AICw',
         by=c('n','q','nXq','rwGsm')[1], round.digits=5)

Arguments

vars

variable names of model

fits

a list of PGLS model fits

optims

a list of PGLS optimization paramters (should include "AICw")

weight

a column name in the optims that specifies the weights to be used in the average

by

unique identifier used to group sub-datasets for reporting (defaults to n)

round.digits

the tnumber of decimal places of the resultant mean to ouput

binary

converts all parameters to binary for presense or absense to calculate 'importance'

standardize

standardize the coefficient estimates by partial standard deviations, according to Cade (2015)

Value

A vector of AICc difference weighted [AICw] averages of PGLS coefficients. Also returns model 'selection' errors or the square root of 'uncertainties' (Burnham & Anderson 2000)

Examples

data.path <- system.file("extdata","primate-example.data.csv", package="mmodely")
data <- read.csv(data.path, row.names=1)
pvs <- names(data[3:5])
data$gn_sp <- rownames(data)

tree.path <- system.file("extdata","primate-springer.2012.tre", package="mmodely")
phyl <- ape::read.tree(tree.path)[[5]]

comp <- comp.data(phylo=phyl, df=data)

mods <- get.model.combos(predictor.vars=pvs, outcome.var='OC', min.q=2)

PGLSi <- pgls.iter(models=mods, phylo=phyl, df=data, k=1,l=1,d=1) 

average.fit.models(vars=c('mass.Kg','group.size'), fits=PGLSi$fits, optims=PGLSi$optim)
variable.importance(vars=c('mass.Kg','group.size'), fits=PGLSi$fits, optims=PGLSi$optim)